[QUARK-479] Fix per-layer quant dtype in DeepSeek attention init#268

Open

thpereir wants to merge 2 commits intothpereir/quark_quant_layerfrom

thpereir/deepseek_r1_mxfp4_ptpc

Commits on Mar 6, 2026

feat: resolve per-layer quant dtype in DeepSeek attention init
thpereir
committed
fix: use per-layer resolve for MTP FP4 bypass in deepseek_mtp
thpereir
committed